<\/p>\n
The editors of Digital Foundry read through the 39-page patent that was made public on New Year’s Eve. According to them, it’s not just a simple Nvidia DLSS definition, as they found a scaling method that they describe. It is expensive in terms of processing power and bandwidth, but here it is defined to do this scaling up in 1\/120 or 1\/160 of a second for a low-power chip. So it’s not exactly the usual DLSS that we see on PC GeForce RTX graphics cards.<\/p>\n
The patent for the AI-based upscaler for the Nintendo Switch 2 also includes a description of what we might call dynamic AI upscaling. According to this description, a model based on real-time data from the game engine can be turned off in real time or pre-programmed to use lower quality upscaling to fit within a certain time budget. Essentially, if the game outputs an image beyond 16.6 milliseconds or 33.3 milliseconds, the device will use a lower quality upscaling model to output the final 1080p image. This is obviously new technology that we haven’t seen in action before, and may have been developed specifically for the Nintendo Switch 2, as it may not be needed for PC desktops (exceptions could be very low-powered laptops).<\/p>\n
By the way, the Nintendo Switch 2 patent is very broad and describes all kinds of applications of AI-based scaling, such as neural decompression, which could be used to reduce file sizes, or even cloud-based scaling, and describes different options, such as scaling in the cloud, then being streamed to a local device, and scaling on the local device.<\/p>\n
The Nintendo Switch 2 is expected to be unveiled by the Japanese company at the end of March. But seeing how many leaks are happening, it could be even sooner.<\/p>\n